Software Defined Sensors Leverage Computational Photography Concepts

Over the past decade, the traditional camera industry has been battered by the consumer shift to using smartphones as their primary image capture device. The fact that a device that is typically less than 10 mm thick can generate an even remotely pleasing image is actually quite astonishing. Some of the thinking that has enabled phones to create what look like well-lit images in near total darkness or portraits with simulated background blur are now being applied to vehicle sensors like lidar. 

Over the past several years, there has been a lot of discussion about the move to software defined vehicles. Increasingly, almost everything in the vehicle is managed by software and individual sensors including lidar are no exception. From controlling the pieces that make the sensor work to processing the signals coming out, software is key to achieving the desired results. As lidar sensors begin to proliferate on new vehicles, engineers are starting to apply some of the concepts of computational photography to improve perception capabilities. 

Most if not all of the coming wave of lidar sensors including those designed by companies like Aeye, Luminar and others have the ability to update their onboard software if the vehicle has over-the-air (OTA) update capability. While most of the lidar that has garnered attention up to this point has focused on medium to long-range sensing capabilities to enable assisted and automated driving systems, a new category of near-infrared light-based sensors is now emerging, near-field lidar (NFL). 

Like most other lidars, NFL sensors utilize light in the near-infrared spectrum with wavelengths of about 905 nm. The light is emitted and the time it takes to reflect back from objects is measured to calculate the distance. The difference between other lidar and NFL is that there is generally no beam steering, they operate at lower power outputs and have lower resolution. NFL sensors from companies such as PreAct Technologies and Valeo have a maximum range of about 15 to 20 meters while high end lidars from Aeye have been demonstrated to detect a vehicle at up to 1,000 m and Luminar sensors can go over 500 m. Those sensors also currently cost up to $1,000. PreAct’s T30P is targeting a cost in high-volume production of $25 or less. 

One of the ways that Aeye, Luminar, Innoviz and others are using software is to control the mechanism that steers the laser beam across the field of view, putting more laser pulses on specific areas of interest and a more sparse point cloud on areas in the periphery.   

PreAct’s low cost, low power sensor is mechanically very simple with a 940 nm LED emitter, a 320×240 pixel photosensor to capture the reflected photons and a processing chip with embedded software. It’s designed to slot into applications that currently use ultrasonic sensors or short-range radar such as parking assist, curb detection, rear automatic emergency braking and imminent crash detection. 

While it doesn’t have a beam steering system to manage, the software can still adjust the operation in real time based on the current context. For example, while most of the previously listed assistance functions are for low-speed operation where short-range is adequate, when the vehicle is moving at higher speeds, with more airflow for cooling, it could increase the power of the emitter for longer range detection, albeit not as long as other more costly sensors. It could also leverage some of the same computational photography techniques used on smartphones, such as stacking multiple frames to get more detail than the sensor might grab in a single frame. Closed loop feedback from the perception software in the driver assistance system can be used in processing the edges of detected objects. 

Like other sensors the PreAct sensor can be expected to improve over time with software updates. As new capabilities are developed, automakers will likely try to find ways to charge customers for new features delivered via OTA updates without having to replace the hardware. 

PreAct claims to have two design wins from automakers for its near field lidar sensors and they are expected to launch on vehicles in mid-decade. PreAct is also offering the sensors for various aftermarket applications with the first product due to launch before the end of 2022. PreAct isn’t saying exactly what sort of product this will be, but CEO Paul Drysch acknowledged that systems like blindspot monitors for trailers would be one potential application.

https://www.forbes.com/sites/samabuelsamid/2022/02/24/software-defined-sensors-leverage-computational-photography-concepts/